9004 Problems of information security. Computer systems Проблемы информационной безопасности. Компьютерные системы 2071-8217 18 10.48612/jisp/npx5-z2hr-h5ux Decentralized approach to intrusion detection in dynamic networks of the Internet of Things basing on multi-agent reinforcement learning and inter-agent communication Децентрализованный подход к обнаружению вторжений в динамических сетях Интернета вещей на базе многоагентного обучения с подкреплением и межагентным взаимодействием 0000-0002-9732-0099 Kalinin Maxim max@ibks.spbstu.ru Tkacheva Ekaterina tkacheva.ki@yandex.ru Peter the Great St. Petersburg Polytechnic University 08 06 2023 2 202 211

The paper proposes a multi-agent reinforcement learning technology for intrusion detection in the Internet of Things. Three models of a multi-agent intrusion detection system have been implemented - a decentralized system, a system with the transmission of forecasts, a system with the transmission of observations. The obtained experimental results have been compared with the open intrusion detection system Suricata. It has been demonstrated that the proposed architectures of multi-agent systems are free from the weaknesses found in the usual solutions

 agent decentralized system internet of things greedy algorithm cybersecurity machine learning multi-agent reinforcement learning intrusion detection observation data transferring prediction data transferring DQN